Agent Lab — a live forecast exhibit · Applied AI
Public track record

How the forecast moved

Every refresh appends a dated snapshot of the model's champion probabilities. This is the accumulating record — P(champion) over time as results land and the field narrows — and, once matches are played, a running score of how well the per-match forecasts held up. Nothing here is back-filled: each point was published at the time shown.

P(champion) over time

31 snapshots

Latest snapshot · movement since first recorded

Team P(champion) Δ since start

Forecast accuracy

Matches scored
4
Top-pick accuracy
75.0%
Mean Brier
0.5798

Multiclass W/D/L · lower is better · 0 = perfect